TheGrebs.com:IRC Logs:#xen:2005:04:12

---	Log	opened Tue Apr 12 00:00:16 2005
01:21	---	---> [e]num [~enum@cpe-24-175-6-109.houston.res.rr.com] has joined #xen
01:22	[e]num	hey.. anyone noticed files starting to come up missing in any of thier domUs'?
01:23	[e]num	I am running an apache host on a domU. Uploaded a template.. and today some of the template files are missing.. css, image files
01:23	[e]num	especially image files
01:31	[e]num	ha.. dom0 disk was full
01:36	---	---> matta-lt [~matta@69.93.28.254] has joined #xen
01:41	---	---> Shaun [~ndci@ip68-111-70-41.oc.oc.cox.net] has joined #xen
02:23	---	---> cfreak [cfreak@dsl-084-056-098-175.arcor-ip.net] has joined #xen
02:24	cfreak	good morning
02:24	cfreak	Apr 12 09:24:00 test kernel: printk: 4 messages suppressed.
02:24	cfreak	Apr 12 09:24:00 test kernel: Memory squeeze in netback driver.
02:24	cfreak	Apr 12 09:24:05 test kernel: printk: 4 messages suppressed.
02:24	cfreak	Apr 12 09:24:05 test kernel: Memory squeeze in netback driver.
02:25	cfreak	xen-testing-src
02:35	---	<<-- wenchien [~wenchien@221-169-69-23.adsl.static.seed.net.tw] has quit (Ping timeout: 480 seconds)
02:44	---	---> jesse_ [~wenchien@221-169-69-23.adsl.static.seed.net.tw] has joined #xen
02:44	---	User: *** jesse_ is now known as wenchien
03:36	---	---> tierra [~tierra@c-24-10-173-249.hsd1.ut.comcast.net] has joined #xen
03:53	---	---> athomas [~athomas@ppp-0-173.lond-b-1.access.uk.tiscali.com] has joined #xen
04:09	---	Netsplit electron.oftc.net <-> iridium.oftc.net quits: xai, jonmason, niv, Shaun, nextime, soffi, caker, Tv, matta-lt
04:09	---	Netsplit electron.oftc.net <-> iridium.oftc.net quits: knewt, @anticw, JViz, plars, lilo, dst, joink, tab, muli_, VS_ChanLog, (+3 more, use /NETSPLIT to show all of them)
04:09	---	Netsplit electron.oftc.net <-> iridium.oftc.net quits: DEac-, tierra, athomas, demon, @surriel, schweeb, Nigelenki, aliguori-hm, matta, unriel, (+5 more, use /NETSPLIT to show all of them)
04:10	mael	mmmh looks like a split
04:13	cfreak	:)
04:13	---	Netsplit over, joins: tierra, Shaun, sunny, schweeb, unriel, drbyte, lilo, @surriel, DEac-, nextime (+16 more)
04:13	---	Netsplit over, joins: caker, Tv
04:18	mael	aaah
04:18	mael	we're back
05:27	---	<<-- cfreak [cfreak@dsl-084-056-098-175.arcor-ip.net] has quit (Quit: .)
05:28	---	---> cfreak [cfreak@84.56.98.175] has joined #xen
06:16	---	---> dst [~dst@p4b23e3d4.np.schlund.de] has joined #xen
06:16	---	---> tab [~tab@darwin.snarc.org] has joined #xen
06:16	---	---> Arnald [~Arnald@81-86-116-102.dsl.pipex.com] has joined #xen
06:16	---	---> matta-lt [~matta@69.93.28.254] has joined #xen
06:16	---	---> joink [~joink@186.80-202-132.nextgentel.com] has joined #xen
06:16	---	---> plars [~plars@pixpat.austin.ibm.com] has joined #xen
06:23	---	---> VS_ChanLog [~stats@ns.theshore.net] has joined #xen
06:23	---	Netsplit uranium.oftc.net <-> kinetic.oftc.net quits: VS_ChanLog
06:23	---	Netsplit over, joins: VS_ChanLog
06:23	---	---> muli_ [~muli@nesher3.haifa.il.ibm.com] has joined #xen
06:38	---	---> jimix [~jimix@ip13.194.susc.suscom.net] has joined #xen
07:27	---	<<-- cfreak [cfreak@84.56.98.175] has quit (Quit: .)
07:27	---	---> cfreak [cfreak@dsl-084-056-098-175.arcor-ip.net] has joined #xen
07:30	---	---> cfreak\| [cfreak@dsl-084-056-098-175.arcor-ip.net] has joined #xen
07:30	---	<<-- cfreak\| [cfreak@dsl-084-056-098-175.arcor-ip.net] has quit (Quit: )
07:58	---	---> harry [~harry@blueice4n1.uk.ibm.com] has joined #xen
08:02	---	<<-- harry [~harry@blueice4n1.uk.ibm.com] has quit (Read error: Connection reset by peer)
08:12	---	---> ville- [~user@dsl-kpogw4a72.dial.inet.fi] has joined #xen
08:15	---	---> mdday [~mdday@bi01p1.nc.us.ibm.com] has joined #xen
08:24	---	<<-- jimix [~jimix@ip13.194.susc.suscom.net] has quit (Read error: Connection reset by peer)
08:35	---	User: *** unriel is now known as riel
08:45	---	<<-- DEac- [~deac@xdsl-84-44-151-215.netcologne.de] has quit (Read error: Operation timed out)
08:47	---	---> hebutterworth [~harry@blueice4n1.uk.ibm.com] has joined #xen
08:56	---	---> hollis [~hollis@user-0vvde2g.cable.mindspring.com] has joined #xen
08:58	---	---> DEac- [~deac@xdsl-213-196-198-171.netcologne.de] has joined #xen
09:23	---	---> pbenner [~pbenner@gate.emlix.com] has joined #xen
09:23	pbenner	hi
09:23	---	User: *** pbenner is now known as philipp
09:25	philipp	Eterm: Error: grantpt(4) failed: No such file or directory
09:25	philipp	Eterm: Error: Can't open pseudo-tty -- No such file or directory
09:25	philipp	Eterm: Error: Unable to run sub-command.
09:25	philipp	how can I create some?
09:26	philipp	xterm: Error 32, errno 2: No such file or directory
09:26	philipp	Reason: get_pty: not enough ptys
09:37	philipp	ah, had to mount devpts
09:37	---	<<-- philipp [~pbenner@gate.emlix.com] has quit (Quit: leaving)
09:50	---	<<-- aliguori-hm [~anthony@cpe-70-112-81-91.austin.res.rr.com] has quit (Read error: Operation timed out)
10:00	---	---> soffi_ [~soffi@Tolvudeild-SC-gw.vv.simnet.is] has joined #xen
10:16	---	<<-- hollis [~hollis@user-0vvde2g.cable.mindspring.com] has quit (Quit: leaving)
10:21	---	---> monrad [~monrad@213083190130.sonofon.dk] has joined #xen
10:24	---	---> jeroney [~jeroney@pixpat.austin.ibm.com] has joined #xen
10:35	soffi_	allright some dude is gonna upload me some Ubuntu and FreeBSD domU images
10:35	mael	cool
10:35	mael	:)
10:35	mael	btw you didn't de-compress the top level archive :)
10:38	soffi_	what?
10:38	soffi_	I just woke up please talk slowly
10:38	mael	hehe
10:38	mael	there's a top level tar.gz
10:38	soffi_	please point
10:38	mael	containing 2 archives (one tar.gz, one tar.bz2, a README and a md5sums file)
10:38	soffi_	My eyes are sore
10:39	soffi_	ahhhhhh xenzoo
10:39	mael	yep :)
10:39	soffi_	uno momento senjor
10:40	murb	I have a nearly working kfreebsd image
10:40	soffi_	ok It's decompressed now senjor
10:43	soffi_	murb: cool
10:43	---	---> hollis [~hollis@pixpat.austin.ibm.com] has joined #xen
10:43	soffi_	mael: I wonder how subversion handles big tarballs ;)
10:43	mael	very badly I guess
10:43	soffi_	ghehe
10:44	mael	is there any cambridge based guy on the chan at the moment?
10:44		* mael need hostelling advices in Cambridge
10:48	soffi_	Cambridge, MA ?
10:48	mael	Cambridge UK
10:48	soffi_	ah
10:49	mael	It seems that am.ac.uk might be in Cambridge :)
11:20	soffi_	mael: have you worked with iscsi or drbd ?
11:21	mael	a friend of mine has worked out a cluster with a drdb raid 1 volume
11:22	mael	But I have no operationnal knowledge of drdb nor iscsi
11:23	soffi_	yeah.. I'm looking into iscsi on two storage servers.. so one of them can fail and nothing goes havoc
11:24	mael	mmh
11:24	mael	not sure of what you mean
11:24	mael	but if you want to share a volume on a server with iscsi, if it fails, it fails
11:24	soffi_	Two storage servers, both serving as iscsi targets, replication between them
11:25	soffi_	only one is used at a time but the other one has up to the millisecond replicas of the primary one
11:26	mael	k then it's an active/passive cluster configuration
11:27	soffi_	yep
11:27	soffi_	just not sure how to do the replication thingie
11:27	mael	it's what my friend had done with drbd
11:28	mael	heartbeat and drbd more precisely
11:29	soffi_	So I could you drdb to replicate the two, and put iscsi on top ?
11:29	soffi_	or am I smoking crack
11:30	mael	you can throw you iscsi stuff :)
11:30	mael	and the crack
11:30	mael	:)
11:30	soffi_	hehe
11:31	soffi_	but iscsi is so cool
11:31	mael	mmh
11:31	mael	if you say so
11:31	mael	encapsulation overhead seems a killer for me
11:32	soffi_	yeah but iscsi is so interoperable
11:32	soffi_	and you can always shell out for an offloading adapter
11:32	mael	and I don't like the idea of mixing data network and access network
11:33	soffi_	No i would but it on a different subnet
11:33	mael	I like the protocol barrier in fibre channel
11:34	mael	I feel more confortable with the idea of cleanly separating public access (internet clients for instance) and the precious data backend
11:35	soffi_	yeah.. with seperate NICs and switches :)
11:35	mael	there's no mistake possible such as a bad route, wrong firewall rules and so on, allowing everyone to read/write your vital data
11:35	mael	it's still the same protocols
11:35	mael	mistakes are always possible
11:36	mael	it gets even worse when you have a lot of clients dedicated networks
11:36	mael	then firewall rules and routing turn into a nightmare
11:37	soffi_	hehehe youre making this too complex :)
11:38	mael	well I used to work for a company where stuff had to be separated very efficiently
11:38	soffi_	hehe
11:38	mael	and in the same time we had to change architecture very often to adapt to user/client demand
11:38	mael	iscsi would have been a nightmare over ther
11:39	mael	+e
11:39	mael	nfs was already
11:50	---	---> _MarkW [~MarkWilli@hebble.cl.cam.ac.uk] has joined #xen
11:54	mael	hi _MarkW
11:54	_MarkW	mael: Hi
11:54	_MarkW	What's up?
11:54	mael	I was waiting for you :)
11:54	mael	well one of you more precisely
11:55	_MarkW	Ah :-)
11:55	_MarkW	It's usually me
11:55	mael	do you know Cambridge well enough to recommand so cheap accomodation there?
11:55	mael	s/so/some/
11:55	_MarkW	errm not exactly
11:56	mael	you don't live there?
11:56	_MarkW	Yes but I've never been a visitor here.
11:56	mael	hehe yeah
11:56	_MarkW	You here on business?
11:56	mael	my GF had an interview for a job there next week
11:57	mael	and we look for a place to stay
11:57	_MarkW	Ah, I see. Tried asking the company if they can recommend anywhere?
11:57	mael	we found a youth hostel in Harlow but I tried to see if you guys had a better idea
11:57	_MarkW	There are quite a lot of Bed and Breakfasts in Cambridge.
11:58	mael	well I think we will book the YH then
11:58	mael	yeah but hard to find when you're abroad
11:58	mael	she's in Germany and me in France
11:58	mael	it's not so easy :)
11:58	_MarkW	sure
11:58	_MarkW	seen this: http://www.touristnetuk.com/em/cambridge/accommodation/ac-serviced/index.htm ?
11:59	_MarkW	google is your friend
11:59	mael	hehe
11:59	mael	on top of that we need a room for three
11:59	_MarkW	the YH will likely be cheaper though but it'll be further away
11:59	mael	yeah but we'll move to London on saturday anyway
11:59	mael	so it's on the road
11:59	_MarkW	ah, ok
12:00	mael	and it's a not-so-expensive solution
12:00	_MarkW	yup
12:01	_MarkW	things that are actually in cambridge are almost undoubtedly going to be more expensive - property prices in cambridge are high, so everyone will be charging more
12:01	mael	I guess so
12:02	mael	it seems to be quite an expensive town
12:02	mael	is the university in the city centre?
12:02	soffi_	go to Hull, it's cheap ;)
12:02	mael	I can offer you a drink if you don't mind a real chat :)
12:03	_MarkW	mael: the university is spread out all over the town
12:03	mael	uh
12:03	mael	soffi: where's Hull?
12:03	soffi_	on the Humberside
12:03	_MarkW	the university is the town :-)
12:03	mael	it's like oxford then?
12:03	soffi_	on the northcoast of england :)
12:04	_MarkW	mael: yeah, like oxford but smaller.
12:04	_MarkW	much smaller.
12:04	mael	I've never been in Cambridge but I visited oxford once
12:05	_MarkW	OK, it'll be familiar then. The university was founded by rebels who had run away from Oxford.
12:05	_MarkW	So they did everything the same :-)
12:05	mael	hehe
12:05	mael	:)
12:06	_MarkW	mael: btw, are you on the mailing list? would i have seen you there?
12:07	mael	I don't remember having posted anything
12:07	mael	I'm in my "read all of the f*g manual first" period
12:08	mael	I still have a hard time to figure out xen's architecture changes
12:08	_MarkW	changes relative to what?
12:08	mael	the frontend/backend part especially
12:09	mael	schemas changed slightly in Ian's presentations
12:09	_MarkW	which presentations are you looking at?
12:10	mael	sorry I was not clear :)
12:10	---	---> niv [~nivedita@bi01p1.co.us.ibm.com] has joined #xen
12:10	mael	OLS 2004 and LW 2005
12:10	mael	on the first one there's a drawing showing vanilla drivers on the privileged domain
12:10	_MarkW	OK
12:11	mael	and frontend devices connecting straight to the vanilla driver
12:11	mael	then on LW the frontend/backend/vanilla model appeared
12:11	_MarkW	right.
12:11	mael	I was not able to dig the change in the ML archives
12:11	_MarkW	let me just look...
12:12	mael	I guess this is all explained in the document i'm ready actually in my commuting time : the one related to safe hardware access
12:12	_MarkW	ian's just missed out the backend drivers in the OLS presentation for clarity purposes
12:13	mael	mmh ok
12:13	_MarkW	backend drivers have been there ever 2.0.
12:13	_MarkW	so the architecture hasn't changed in the timeframe you're looking at, just the diagrams ;-)
12:13	---	<<-- athomas [~athomas@ppp-0-173.lond-b-1.access.uk.tiscali.com] has quit (Quit: Leaving)
12:13	mael	fact is the presentations (text-less) are one of the only documents with xen architecture
12:14	mael	I'm trying to figure out things, but as I'm not a native C-speaker, I read the documentation before the source code
12:14	mael	(my C level is far beyond my english)
12:15	_MarkW	mael: sure. the paper you're reading is also good material
12:15	mael	you mean, apart from the fact you got your name on it? :))
12:16	_MarkW	yeah, apart from that :-)
12:17	_MarkW	i worked on the driver isolation stuff for that paper
12:18	mael	a prototype to prove the stuff?
12:19	mael	I guess the prototype ended in xen production code anyway
12:19	_MarkW	well, it all works - it's in the 2.0.x tree at the moment
12:19	_MarkW	but it could do with being a bit more user friendly
12:19	mael	there's something I don't get also
12:19	_MarkW	most people prefer to run drivers in dom0, in which case you don't get any more isolation than vanilla linux
12:19	_MarkW	oh?
12:20	mael	uh? what do you call isolation in that case?
12:20	mael	error trapping and the like?
12:20	mael	stuff you're talking about in the paper?
12:20	_MarkW	yup
12:21	mael	ok I'll read it first then, and ask after :)
12:21	_MarkW	making drivers less privileged so that driver bugs are less able to crash your machine
12:21	_MarkW	cool :-)
12:21	mael	hehe
12:23	mael	yeah the other stuff was related to the changes needed in the hypervisor in order to support different hw architecture
12:24	mael	x86-64 of course (the xen 3.0 summit talks about it), but also ia64
12:24	---	<<-- niv [~nivedita@bi01p1.co.us.ibm.com] has quit (Ping timeout: 480 seconds)
12:24	mael	it seems that the hypervisor will need to be separated in multiple layers
12:25	knewt	gah to people who send html-only emails. i wish the list would just reject such messages
12:25	mael	one for arch specific treatments (to do strong VM separation), one for ressources access/scheduling and the like, and maybe some more I don't get
12:25	_MarkW	mael: there's xen/common and xen/arch for splitting out this stuff
12:26	_MarkW	the interfaces between the two are still evolving
12:26	soffi_	hmmm ATA over Ethernet
12:26	_MarkW	but there will be a lot of arch-dependent code - far higher proportion than in Linux
12:27	mael	_MarkW: yeah, this is what I'm thinking about
12:27	mael	what would be an approx. percentage of common/relative code?
12:29	mael	I fear this would be the usual 30/70
12:29	mael	or even worse
12:29	riel	well, the architecture dependant code would be very similar to the non-xen arch code
12:29	eigood	AoE sucks
12:29	eigood	very slow
12:30	mael	riel: oh yeah?
12:31	mael	you mean similar in volume or in design?
12:32	mael	maybe both
12:32	hollis	depends on what is done in dom0 and what isn't
12:32	soffi_	eigood: how slow? :)
12:34	eigood	5MB/s on a 100M lan
12:34	eigood	that's on native linux; athlon 2000 client, p2-2.4g server
12:34	eigood	p4-2.4g
12:34	riel	mael: lots of it would be similar in design
12:35	riel	mael: in fact, it already is
12:35	mael	for x86 you mean?
12:35	riel	for x86-64 too
12:35	mael	oh
12:35	hollis	well, for IA64 they literally copy in great gobs of Linux code
12:35	mael	ok
12:35	mael	this is not such a waste then :)
12:36	hollis	what isn't?
12:38	mael	I feared the project was too x86-centric
12:38	hollis	well it is :)
12:38	soffi_	man that is slow
12:39	mael	as for me the barriers xen's paravirtualisation solution where trying to break are x86 specific
12:39	mael	at least this is what I understood
12:39	riel	hollis: not necessarily a bad thing - x86 and x86-64 are 95% of the PC class computers ;)
12:39	riel	with PPC being most of the other 5%
12:39	hollis	riel: luckily for us, Linus took a different attitude towards architecture portability...
12:40	hollis	mael: most architectures have very similar problems they need to solve
12:40	riel	hollis: "portability" isn't the same as "focus on something non-common"
12:41	mael	hollis: ok
12:41	hollis	sure, but there seems to be limited interest even in portability at the moment
12:41	hollis	riel: I'm not asking anybody to make PPC work for me... :)
12:41	mael	but even though they are similar, dealing with them may need a huge amount of effort to deal with the specificities
12:42	soffi_	but one question.. will it ever be ported to Amiga?
12:42	soffi_	:-P
12:42	mael	;)
12:43	riel	just run multiple amiga emulators ;)
12:43	hollis	mael: yes. but at the same time, there are some aspects of hypervisors (such as the virtual IO drivers, NUMA awareness, resource management, scheduling, console, ...) that can be shared across those disparate architectures
12:43	soffi_	hehehe
12:43	mael	hollis: yup
12:43	mael	but what is the relative amount of code/effort?
12:44	mael	this is my concern in the first place
12:44	hollis	relative to what?
12:44	mael	let's say amount of code produced/modified
12:45	mael	for both aspects
12:45	hollis	lines of code? for what? I'm sorry, I'm not understanding your question...
12:45	mael	sorry
12:45		* mael is trying to reformulate
12:46	mael	how much of the code produced by xen project on x86 architecture will be usable when porting to another arch?
12:47	riel	do you refer to code written specifically for xen, or also to x86-specific code that was slightly modified to accomodate xen ?
12:47	mael	both
12:48	riel	then you may want to rethink your question ;)
12:48	mael	hehe
12:48	riel	surely it would be easier to slightly modify eg. PPC code, instead of porting x86-specific code with Xen modifications over to PPC ?
12:48	soffi_	I'll trade you 5 tons of snow for 5 minutes of sunshine
12:49	mael	riel: yeah sure but I was not trying to ask that :)
12:49		* mikegrb slightly modifies riel
12:49	riel	mael: then don't say "both" ;)
12:49	mael	uh?
12:49	mael	ok then
12:49	hollis	mael: all of those things I listed before can be usable when porting to another architecture
12:50	riel	it should be easy to "port" the non-x86 specific bits of Xen over to other architectures, since they are already non-arch-specific
12:50	riel	maybe some generalisations will be needed, but certainly no port
12:50	hollis	yeah
12:50	mael	hollis: yes and I guess this is what _MarkW was calling the xen/common stuff
12:51	hollis	mael: yup
12:51		* _MarkW grunts and wakes up
12:51	mael	but what about the xen/arch code?
12:51	hollis	mael: the problem is that since those guys are writing both the common and arch code, the interface can get a little blurred sometimes
12:51	mael	this is kind of wasted when you address arch specific problems, no?
12:52	mael	ooh
12:52	mael	okay
12:52	hollis	right, I am not porting xen/arch/x86 to PPC, but rather creating a new xen/arch/ppc64
12:52	mael	okay, then we're coming to my question :)
12:53	---	<<-- matta-lt [~matta@69.93.28.254] has quit (Quit: Why is the alphabet in that order? Is it because of that song?)
12:54	mael	the code you produce for the xen/ppc64 arch will only be used for xen, right?
12:55	hollis	yes, though some of it comes from Linux and some of it comes from rhype (IBM Research Hypervisor)
12:55	_MarkW	mael: there common xen/arch stuff between x86 and x86_64. common stuff in XenLinux for multiple archs should probably go in drivers/xen
12:55	mael	_MarkW: ok
12:56	_MarkW	since arch/xen is getting pulled into arch/i386, there's nowhere else to put the common code :-)
12:57	hollis	_MarkW: hang on, I was talking about the Xen core, not the Linux patches
12:57	_MarkW	hollis:i thought we'd moved onto talking about Linux stuff now... saw the phrase "arch/xen" and assumed
12:57	hollis	mael: you were talking about the Xen core, yes?
12:58	_MarkW	ok, so in Xen I guess the plan is to make the interfaces clean and maximise the amount of stuff in xen/common that the arch ports can use
12:58	mael	mmmh I am very tempted to say "both", actually
12:58	mael	:)
12:58	_MarkW	in XenLinux, the plan is the similar but with common stuff going into drivers/xen instead
12:59	_MarkW	in XenLinux, I think the plan is to incorporate Xen support into each Linux architecture tree (i386, ppc64, etc)
12:59	---	<<-- soffi_ [~soffi@Tolvudeild-SC-gw.vv.simnet.is] has quit (Read error: Connection reset by peer)
13:00	---	---> soffi_ [~soffi@Tolvudeild-SC-gw.vv.simnet.is] has joined #xen
13:00	mael	what I meant in the first place was : is it a long/difficult process to paravirtualize (=adapt to xen hypervisor design) an arch? and if so, is it efficient to use xen then, instead of writing a new hypervisor from scratch?
13:00	hollis	mael: PPC is a special case, because PPC Linux has run on IBM's hypervisors for years
13:00	mael	I guess answears will be something like "that reinventing the wheel is always a waste"
13:01	mael	hollis: ok
13:01	mael	-that
13:01	hollis	in some sense you could see xen/ppc as "reinventing the wheel" I guess
13:02	mael	well I guess it has other advantages
13:02	mael	(I mean xen vs. ibm's hypervisors)
13:02	riel	migration
13:02	hollis	for example, IBM could never open source the "real" hypervisor found on pSeries
13:03	riel	common management tools
13:03	riel	being fully open source
13:03	riel	there are probably more advantages
13:04	mael	yeah, even though a "common vmm management interface specification" would have some of the advantages you spotted
13:04	mael	but you're right
13:06	mael	it's just things are not very clear for me sometimes
13:06	hollis	then you are in good company I think :)
13:06	mael	yeah :)
13:07	mael	I was confused by the L4onLinux guy the other day
13:07	_MarkW	mael: sometimes things i've done aren't even clear to me - why did I write that line of code? why did i put all my food? that sort of thing.
13:08	_MarkW	mael: which guy?
13:08	mael	_MarkW: hold on I'm trying to localize him in the logs :)
13:11	mael	it was april the 07
13:13	mael	his nick was rtc2 and he asked how xen to be compared to L4Linux
13:15	mael	L4Linux is (from what I understood after a quick reading of their website) a porting of linux kernel on top of a L4 microkernel
13:15	mael	website is http://os.inf.tu-dresden.de/L4/LinuxOnL4/
13:15	hollis	it's a good question
13:15	---	---> niv [~nivedita@bi01p1.co.us.ibm.com] has joined #xen
13:21	mael	since then I decided to try do have a more indepth look at xen arch, as I suspected most of the differences to be at the low level layers
13:21	mael	even if from a non-technical user point of view things might appear somewhat similar
13:22	hollis	if you don't know C it might be difficult to compare the two
13:23	mael	well I don't think you have to read all the sources to figure out the differences in the concepts
13:24	knewt	omfg. the US army now has WLAN-controlled landmines. and is going to be deploying them in iraq.
13:25	mael	ouch! already 8:30pm...
13:25	mael	time to read the safe hardware access paper in the train!!
13:25	mael	see you
13:26	riel	knewt: interesting, the army's own computer safety guidelines basically prohibit WLAN on their equipment
13:27	mael	riel: they probably explode when you're around with your wlan equipment :)
13:27	knewt	http://www.theregister.co.uk/2005/04/12/laptop_triggered_landmine/
13:32	---	<<-- hebutterworth [~harry@blueice4n1.uk.ibm.com] has quit (Quit: Leaving)
13:37	---	<<-- soffi_ [~soffi@Tolvudeild-SC-gw.vv.simnet.is] has quit (Quit: Leaving)
13:48	[e]num	yo.. when building xen, is it possible to force a --prefix?
13:48	[e]num	since I know there is no ./configure script
13:51	---	User: *** anticw is now known as cw
13:54	knewt	[e]num: assuming you're talking about the tools bit, just specify a DESTDIR when doing the install step
13:55	knewt	oh, wait, that probably won't do exactly what you want
13:55	aliguori	[e]num: no, Xen assumes certain paths
14:02	---	<<-- niv [~nivedita@bi01p1.co.us.ibm.com] has quit (Quit: Quitting)
14:08	---	<<-- cfreak [cfreak@dsl-084-056-098-175.arcor-ip.net] has quit (Quit: .)
14:13	---	---> niv [~nivedita@bi01p1.co.us.ibm.com] has joined #xen
14:15	---	<<-- DEac- [~deac@xdsl-213-196-198-171.netcologne.de] has quit (Quit: Verlassend)
14:15	---	---> DEac- [~deac@xdsl-213-196-198-171.netcologne.de] has joined #xen
14:16	eigood	add a top-level Makefile.vars, have each makefile include it, then modify the paths to reference a $(prefix)
14:16	eigood	then, you can just edit that file, specify a different prefix on the cmdline, or at some point if it ever gets a configure script, it can be generated
14:16	---	<<-- DEac- [~deac@xdsl-213-196-198-171.netcologne.de] has quit (Quit: )
14:20	---	---> DEac- [~deac@xdsl-213-196-196-16.netcologne.de] has joined #xen
14:24	---	<<-- DEac- [~deac@xdsl-213-196-196-16.netcologne.de] has quit (Quit: )
14:30	[e]num	yo.. I got this during boot
14:30	[e]num	WARNING: Currently emulating unsupported memory accesses
14:30	[e]num	** in /lib/tls libraries.
14:30	[e]num	I dont have a /lib/tls any ideas?
14:30	riel	that's ok
14:30	riel	probably some other program trying to use TLS
14:30	[e]num	ah
14:30	riel	no need to worry since TLS emulation is correct
14:30	riel	it's just a tad slow
14:30	[e]num	dont know what would use it since I did an LFS.. but as long as it functions as it should
14:31	riel	I only run with TLS
14:31	riel	(since I want to verify that TLS emulation never breaks)
14:31	[e]num	heh
14:31		* riel tries to stay as much on the leading edge as possible
14:31	riel	if something breaks, I can fix it before users find out there was a problem
14:32	eigood	it'd be cool if when xen detected that, it could ask what the dom# was doing at the time, so it could print out what program was running
14:34	---	---> DEac- [~deac@xdsl-213-196-196-194.netcologne.de] has joined #xen
14:34	riel	actually, I think the domain is already notified
14:34	riel	chances are it's just one printk line extra to point out the program
14:34	---	<<-- xai [~pasta@cpe-70-112-17-10.austin.res.rr.com] has quit (Ping timeout: 480 seconds)
14:35	[e]num	hmm.. is there a way to get output from xend start
14:36	[e]num	it seems to load, but when I run xm list it says it isnt running
14:36	aliguori	[e]num: chances are you have to rm -rf /var/lib/xen/*
14:36	jeroney	[e]num: check /var/log/xend-debug.log
14:37	[e]num	OSError: [Errno 2] No such file or directory: '/var/lib/xen/xend-db/domain/'
14:38	[e]num	that doesnt look too god
14:39	[e]num	damn.. this thing is slow as hell
14:39	[e]num	lsmod is taking more than a minute to run
15:23	---	---> tierra\|w_ [~tierra@dsl093-225-126.slc1.dsl.speakeasy.net] has joined #xen
15:25	---	User: *** tierra\|w_ is now known as tierra\|w
15:27	[e]num	hmm.. okay this is interesting..
15:27	[e]num	when I built xen it compiled both dom0 and domU kernels, and placed them in boot
15:28	[e]num	I created a phys vbd, and attempted to boot the domU. the boot goes up intul fsck, and fails saying it cannot find /dev/sda1
15:31	[e]num	does CONFIG_SCSI_SATA have to be configured into the domU kernel?
15:43	eigood	whatever device is being emulated in domU needs to be compiled in domU
15:44	eigood	it's one of those poorly implemented things
16:06	---	---> plars_ [~plars@pixpat.austin.ibm.com] has joined #xen
16:08	---	<<-- plars [~plars@pixpat.austin.ibm.com] has quit (Killed (NickServ command used by plars_))
16:08	---	User: *** plars_ is now known as plars
16:13	---	---> GvG [GvG@geldorp.xs4all.nl] has joined #xen
16:14	_MarkW	eigood: [e]num i didn't think you had to compile in SCSI support to use SCSI drives in a domU
16:14	_MarkW	the VBD driver doesn't talk to the SCSI layer anyhow, so I wouldn't have thought it mattered.
16:14	_MarkW	[e]num: what if you export as /dev/hda1 ?
16:15	_MarkW	mael: you still here?
16:16	_MarkW	mael: L4 was designed from a microkernel mindset, so it provides abstractions that look like minimal kernel features
16:16	_MarkW	e.g. "tasks", "address spaces", "inter process communication"
16:17	_MarkW	Xen differs in that it provides an interface that's more hardware-like: you just get memory, event channels (much like interrupts) and the hypercalls (which are much like CPU privileged instructions)
16:19	[e]num	_MarkW, I found out what it was.. LFS uses udev to populate the dev directory
16:20	[e]num	so I turned udev off, and statically created sda1 with mknod
16:20	[e]num	booted right up after that
16:24	---	---> rharper [~rharper@pixpat.austin.ibm.com] has joined #xen
16:30	_MarkW	[e]num: udev should work fine anyhow
16:30	_MarkW	[e]num: but it'll break if you don't use an initrd
16:33	demon	well, only if you use FC3's udev setup, where it completely eliminates all real nodes in /dev
16:33	demon	with at least some other distros, there are real nodes there so that the system can get off the ground
16:34	_MarkW	demon: ok. i don't know what LFS does - never used it.
16:35	demon	ouch... LFS
16:35	aliguori	yeah, gentoo actually saves the contents of /dev across reboots.. it's quite now
16:39	_MarkW	ok. i assumed everyone kept /dev empty like FC.
16:39	_MarkW	i live and learn :-)
16:39	aliguori	:-)
16:39	aliguori	_MarkW: how's XenFS coming along?
16:40	riel	_MarkW: btw you don't need an initrd
16:40	aliguori	i've heard whispers that it might make 3.0 :-)
16:40	_MarkW	riel: really? what do you do instead?
16:40	riel	_MarkW: if you just put 'null' and 'console' in the empty /dev directory (with udev not mounted), things will work fine
16:40	riel	_MarkW: this is documented in the FedoraXenQuickstart page ;)
16:40	_MarkW	riel: ah, ok. that should probably go into the Xen FAQ too. or at least we should link there - it seems to be a very common question.
16:41	riel	yeah, you might want to link it ;)
16:41	_MarkW	aliguori: might be in beta by 3.0 but i doubt it'll be production quality
16:41	riel	unfortunately I'm not allowed to edit that page (or at least, I wasn't last I looked) or I would have added it already
16:41	_MarkW	it's going OK
16:42	_MarkW	riel: interesting... well if you can't get there then i probably can't either! i was supposed to acquire admin rights at some stage but i'm not sure that's happened yet
16:42	eigood	it'd be better to include it, not link it
16:42	eigood	I don't think it's best if xen pushes any particular distro
16:42	riel	eigood: and be out of date next time the fedora page is changed ?
16:42	eigood	also considering xen deals with more than just linux
16:42	aliguori	yeah, it's kind of hard to figure out where to add stuff in the wiki as so many of the pages are read-only
16:43	_MarkW	eigood: this is a specifically Linux issue though
16:43	eigood	but not just redhat/fedora
16:43	_MarkW	but I guess the same fix should work for other distros...
16:44	_MarkW	eigood: exactly. we should probably link to distro howtos somewhere but if other distros can make do with just /dev/null and /dev/console then that could go in the generic faq
16:44	aliguori	correct me if i'm wrong riel, but this is not a Xen issue as much as it's a Fedora issue
16:44	aliguori	you probably can't boot Fedora in general without an initrd (unless you had /dev/console and /dev/null)
16:45	aliguori	unless Fedora has a patched kernel...
16:45	_MarkW	riel: that's interesting - somebody has added it to the FAQ page in the Xen wiki...
16:46	_MarkW	it wasn't there last time i looked
16:46	---	<<-- jeroney [~jeroney@pixpat.austin.ibm.com] has quit (Quit: Leaving)
16:46		* _MarkW shrugs
16:52	riel	cool
16:53	riel	aliguori: the xenU kernel has ext3 built-in
16:53	riel	aliguori: so you don't need the initrd
16:53	---	---> jeroney [~jeroney@pixpat.austin.ibm.com] has joined #xen
16:53	_MarkW	aliguori: re XenFS, I was going to implement it a stupid way first
16:54	_MarkW	just to get it working quickly. but i've decided that it's actually easier to implement it the cunning way straight off
16:54	aliguori	_MarkW: have you changed your mind? :-)
16:54	---	---> jbailey [~jbailey@CPE000ded9d787c-CM014260028338.cpe.net.cable.rogers.com] has joined #xen
16:55	_MarkW	if i can get direct page sharing working then coherency is (relatively) trivial
16:55	riel	how can you do direct page sharing when a domain migrates away ? ;)
16:56	_MarkW	riel: that's a problem for later work ;-)
16:56	riel	fair enough
16:56	_MarkW	first step for XenFS is to optimise for sharing within one machine
16:57	aliguori	_MarkW: what about between multiple domains?
16:57	---	<--- jbailey [~jbailey@CPE000ded9d787c-CM014260028338.cpe.net.cable.rogers.com] has left #xen (Leaving)
16:57	_MarkW	exactly how migration will work isn't clear yes
16:57	_MarkW	aliguori: that's what i'm interested in
16:57	_MarkW	first off, coherent multiple writer sharing
16:58	aliguori	how do you achieve that?
16:58	_MarkW	(in fact as a side effect, shared memory between processes in different domains should Just Work(TM) by mmaping a file)
16:59	aliguori	do you have multi-domains mutexes or something? or do you guys have a clever lock-free mechanism :-)
16:59	aliguori	_MarkW: neat!
17:00	_MarkW	the fsfront / fsback drivers will do some gyrations to make the page frame backing a file page be the same page frame in every (participating) page cache
17:01	aliguori	oh, but not for metadata right? so if you use the control channels for metadata locking doesn't matter (assuming you can lock at the inode level)
17:01	_MarkW	the plan is when one domain updates its page cache, the other domains will just see it there (there'll need to be some signalling to make sure changes are noticed and written out by the server)
17:01	riel	_MarkW: http://wiki.xensource.com/xenwiki/WhosDoingWhat is an immutable page, too ;(
17:01	riel	_MarkW: it would be useful if that page was editable by everybody with a login
17:01	riel	oh wait -
17:01	riel	my cookie expired ;)
17:02	---	<<-- niv [~nivedita@bi01p1.co.us.ibm.com] has quit (Quit: Quitting)
17:02	@cw	riel: /~ C is fo Cookie ~/
17:02	hollis	in the UK do they call them biscuits? :)
17:02	riel	J is for join, which nobody uses
17:03	@cw	yeah, and other places
17:03	_MarkW	hollis: i suppose we should really
17:03	@cw	nz, au, sg, probably a few others
17:03	_MarkW	mmmm food
17:03	@cw	I'm bummed they are making Cookie Monster more PC. When I was a kid he was one of the cooler characters.
17:04	aliguori	so what do they call a biscuit in the uk if a cookie is a biscuit?
17:04	riel	cw: what?! they're even taking the fun out of sesame street?
17:04		* riel wonders what this world is turning into
17:04	@cw	riel: cookies are bad, so the song and everything are being ruined
17:04	aliguori	riel: yeah, he'll eat cookies in moderation
17:05	@cw	sigh, every kid gets a packet of cookies and pretends to be cookie monster at some point
17:05	_MarkW	aliguori: biscuits are biscuits ;-) cookies are a subclass of biscuits
17:05	riel	throwing up is the best way to learn that cookies can be bad ;)
17:06	_MarkW	riel: cookies are only bad in the same way that pizza is bad
17:06	_MarkW	i like that kind of bad!
17:06	aliguori	pizza is great!
17:06	_MarkW	aliguori: exactly :-)
17:06	aliguori	i have pizza 4-5 times a week.. and i'm not dead yet
17:06	aliguori	so it's gotta be good :-)
17:06	_MarkW	whoa!
17:07		* _MarkW humble bows to aliguori, the master of pizza
17:07	aliguori	i even worked in a pizza place back in jersey :-) it's in my blood man :-)
17:08	@cw	oscar is cool too, i would so live in a garbage can too if i had to deal with everyone on Sesame Street
17:08	_MarkW	how about living in the garbage can at the cookie factory?
17:09	_MarkW	(if you had your pick of cans, that is)
17:12	_MarkW	then you'd get tons of broken cookies that they couldn't package. also, broken cookies have the calories leak out so they're more healthy.
17:12	riel	you believe the "magic blue smoke" story, too ? ;)
17:12	aliguori	_MarkW: you can also shake out more calories out of the cookie
17:13	_MarkW	riel: that's right. i often let the magic smoke out of things.
17:16	_MarkW	sometimes i let virtual magic smoke out of my xen domains also
17:17	---	<--- mdday [~mdday@bi01p1.nc.us.ibm.com] has left #xen (/dev/stomach)
17:19	hollis	:)
17:23	---	<<-- jeroney [~jeroney@pixpat.austin.ibm.com] has quit (Quit: Leaving)
17:42	knewt	what's the function tidyup_all() mentioned in grant-tables.txt?
17:43	---	<<-- JViz [Anomaly@cpe-024-167-184-225.triad.res.rr.com] has quit (Quit: YES THEY DESERVE TO DIE, AND I HOPE THEY BURN IN HELL!)
17:45	---	---> niv [~nivedita@bi01p1.co.us.ibm.com] has joined #xen
17:46	---	User: *** riel is now known as unriel
17:47	---	<<-- hollis [~hollis@pixpat.austin.ibm.com] has quit (Quit: leaving)
17:56	_MarkW	knewt: haven't seen it
17:56	_MarkW	what's it suposed to be for?
17:57	soffi	damn.. if I could get Grand Theft Auto to run on a xenU I could live-migrate it to and from work
17:57	---	<<-- rharper [~rharper@pixpat.austin.ibm.com] has quit (Quit: Leaving)
17:58	_MarkW	knewt: ah, it looks like a hypothetical function for cleaning up any grants that succeeded before one failed
17:58	knewt	in the example code, after the hypervisor grant_table_op call, there's a loop through the mapped frames, and inside it does this: if ( unlikely(aop[i].u.map_grant_ref.dev_bus_addr == 0) ) { tidyup_all(aop, i); goto panic; }
17:58	_MarkW	implementation left to the reader :-)
17:58	knewt	ah, ok
17:59	_MarkW	it passes the pointer to the array of ops and the current counter, which is probably all you'd need
18:00	knewt	so it just does another grant_table_op, but this time with GNTTABOP_unmap_grant_ref ?
18:00	_MarkW	yeah probably. alternatively it could panic or hire a hitman to shoot the programmer.
18:00	_MarkW	your solution is probably the most useful
18:01	---	<<-- GvG [GvG@geldorp.xs4all.nl] has quit (Quit: Goodnight)
18:05	---	<<-- demon [demon@newcastle.devrandom.net] has quit (Ping timeout: 480 seconds)
18:19	niv	anyone remember if there was an oops in zap_pmd_range recently on domU?
18:29	_MarkW	niv: don't remember seeing one
18:29	_MarkW	-unstable tree?
18:30	niv	_MarkW: yep, but 0401 kernel
18:30	_MarkW	0401?
18:31	niv	April 1 nightly snapshot
18:31	_MarkW	Oh right, OK.
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: Unable to handle kernel paging request at virtual address c6ec35e4
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: printing eip:
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: c01434f6
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: *pde = ma 7ff64067 pa 00019067
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: *pte = ma 079aa063 pa 06ec3063
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c0143718>] zap_pmd_range+0x57/0x75
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c0143770>] zap_pud_range+0x3a/0x5e
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c0143808>] unmap_page_range+0x74/0x8b
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c014391c>] unmap_vmas+0xfd/0x223
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c01489b7>] exit_mmap+0x84/0x156
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c01158ee>] mmput+0x2b/0x94
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c015e899>] exec_mmap+0xcb/0x12b
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c015eaae>] flush_old_exec+0x123/0x9ca
18:31	niv	Apr 12 15:29:48 nivdom1 kernel: [<c015436b>] vfs_read+0xb9/0x129
18:32	niv	Apr 12 15:29:48 nivdom1 kernel: [<c015e7c5>] kernel_read+0x4a/0x53
18:32	niv	Apr 12 15:29:48 nivdom1 kernel: [<c017c627>] load_elf_binary+0x37d/0xc9a
18:32	niv	Apr 12 15:29:48 nivdom1 kernel: [<c013910e>] buffered_rmqueue+0x153/0x2ba
18:32	niv	Apr 12 15:29:48 nivdom1 kernel: [<c02016e9>] copy_from_user+0x3d/0x64
18:32	niv	Apr 12 15:29:48 nivdom1 kernel: [<c017c2aa>] load_elf_binary+0x0/0xc9a
18:32	niv	Apr 12 15:29:48 nivdom1 kernel: [<c015f5ee>] search_binary_handler+0xbb/0x2e9
18:32	niv	brb..
18:32	_MarkW	niv: don't thing i've heard of anything like that recently
18:32	niv	ok..will be right back..thanks..
18:33	_MarkW	ok, just prod me when you're back. it doesn't look very good! are you running anything strange like shadow mode, grant tables on, xens and linuxs built at different times?
18:57	---	---> alex234 [new@pD9E1FCF0.dip.t-dialin.net] has joined #xen
19:02	niv	_MarkW: no, nothing unusual - default config, compiled together, etc...
19:02	_MarkW	that's a bit strange. what were you doing with it?
19:03	niv	it was idle!
19:03	niv	wasn't at the machine, came back to find that on my screen
19:04	niv	Had done a bunch of pings earlier..
19:04	_MarkW	hrrrm, sounds like a gremlin!
19:05	_MarkW	i guess reproducing will be tricky then...
19:06	niv	Yep, can't think of what to try, but am giving it a go - just going to run random tests etc
19:06	niv	anything that I can instrument?
19:06	niv	or should be looking at?
19:07	_MarkW	No idea!
19:07	_MarkW	if you're not doing anything unusual then I really don't know what it could be
19:07	niv	Well, someone else will run into it as well. I do have scsi disks, and running on an SMP
19:08	_MarkW	I guess you could post to xen-devel, maybe one of the others will have a clue.
19:08	_MarkW	SMP guest?
19:08	niv	but not an SMP kernel, UP kernel, but SMP box - not that it should make a difference, but..
19:09	_MarkW	it's good to be paranoid :-)
19:09	niv	:)
19:12	---	<<-- drbyte [~byte@c210-49-121-44.eburwd3.vic.optusnet.com.au] has quit (Read error: Connection reset by peer)
19:15	---	---> drbyte [~byte@c210-49-121-44.eburwd3.vic.optusnet.com.au] has joined #xen
19:21	---	<<-- soffi [~soffi@proxy.du.vdsl.is] has quit (Quit: Leaving)
19:47	---	<<-- alex234 [new@pD9E1FCF0.dip.t-dialin.net] has quit (Read error: Connection reset by peer)
19:53	---	---> alex234 [new@pD9E1FCF0.dip.t-dialin.net] has joined #xen
19:55	---	---> demon [demon@newcastle.devrandom.net] has joined #xen
20:04	---	<<-- alex234 [new@pD9E1FCF0.dip.t-dialin.net] has quit (Quit: Leaving)
20:30		* _MarkW yawns
20:30	_MarkW	enough XenFS for tonight. gnight all.
20:30	---	<<-- _MarkW [~MarkWilli@hebble.cl.cam.ac.uk] has quit (Quit: Kopete 0.9.2 : http://kopete.kde.org)
20:31	---	<<-- Shaun [~ndci@ip68-111-70-41.oc.oc.cox.net] has quit (Read error: Connection reset by peer)
20:34	---	<--- niv [~nivedita@bi01p1.co.us.ibm.com] has left #xen (Leaving)
20:59	---	---> Gollum [~gollum@pcp01411681pcs.phnixv01.pa.comcast.net] has joined #xen
21:00	Gollum	just wondering if there are any irq problems in the latest unstable build...
21:00	Gollum	anyone know anything about that?
21:03	---	---> hena [hena@hack.fi] has joined #xen
21:03	---	---> rusty [~rusty@bh02i525f01.au.ibm.com] has joined #xen
21:08	rusty	aluguori: hi!
21:09	Gollum	hello folks. just wondering if there are any irq problems in the latest unstable build...
21:11	rusty	Hello.
21:17	rusty	aliguori: sorry, will try to spell right.
21:20	---	<<-- tab [~tab@darwin.snarc.org] has quit (Ping timeout: 48